t0ddharris Granola Meeting Notes
Pull meeting notes from Granola's API via the CLI wrapper at .claude/skills/granola/granola.sh.
Setup
The user needs a Granola API key (requires Business or Enterprise plan):
- Open Granola desktop app > Settings > API > "Create new key"
- Add it to the project
.envfile:GRANOLA_API_KEY=grn_your_key_here
If the key isn't configured, the script will tell the user what to do.
Commands
SCRIPT=".claude/skills/granola/granola.sh"
# List recent notes (last 7 days)
$SCRIPT recent
# List notes from last N days
$SCRIPT recent 30
# List with date filters
$SCRIPT list --after 2026-04-01 --before 2026-04-07
# Get a specific note (summary only)
$SCRIPT get not_ABC123DEF456GH
# Get a note with full transcript
$SCRIPT get not_ABC123DEF456GH --transcript
# Pull a note into /incoming/ (includes transcript by default)
$SCRIPT pull not_ABC123DEF456GH
# Get raw JSON output (for any command)
$SCRIPT list --json
$SCRIPT get not_ABC123DEF456GH --json
Workflow
When the user asks to pull meeting notes:
1. List available notes
Run recent or list with appropriate date filters to show what's available. Let the user pick which meeting(s) they want.
2. Fetch the note
Use get with --transcript to see the full content. Review the summary and transcript.
3. Decide where it goes
Based on the meeting content, route it appropriately:
| Content type | Destination |
|---|---|
| Customer call, demo, sales conversation | pull to /incoming/, then mine for brain file updates, case study material, or audience language |
| Product feedback, feature requests | Extract insights, suggest updates to /brain/truth.md or /brain/positioning-and-messaging.md |
| Internal strategy discussion | Summarize key decisions, suggest project memory updates |
| Partner or analyst meeting | Extract relevant intelligence for /brain/market-signals.md |
| Content planning meeting | Extract ideas for /marketing/inspiration/content-ideas.md |
4. Process the notes
Apply the same rigor used for Apollo transcript mining:
- Interpret through context (meeting notes may have transcription errors)
- Extract actual customer/prospect language for
/brain/audience-language.md - Identify conversion triggers and objections for
/brain/customer-journey.md - Flag any product claims that need verification against
/brain/truth.md - Never parrot raw transcript, always synthesize
Important Notes
- The API is read-only and returns only notes with AI-generated summaries
- No search endpoint exists; use date filters to narrow results, then review locally
- Page size max is 30; use
--cursorfor pagination - Rate limit: 5 requests/second (burst 25). Don't hammer it
- Transcript speakers are labeled "You" (microphone) and "Other" (speaker audio); actual names come from the attendees list
- The
/incoming/folder is the user's drop zone;pullwrites there by convention so the standard intake workflow applies